by Kimberly Nevala

Making

of

Artificial Intelligence

Making Sense of AI 1 2 3 4 5 2

TODAY’S ARTIFICIAL INTELLIGENCE (AI) SOLUTIONS are not sentient in the manner popularized in science fiction by scores of self-aware and typically nefarious androids. Even so, the ability to arm such systems with the ability to directly sense and respond to their in situ environment is critical.

Why? In the future, our experiences will be smart, intuitive and informed by analytics that are not seen but felt via new business, personal and operational engagement models. Enabling this interaction requires AI applications that can sense, analyze and respond to their environment in an intelligent and interactive manner. Without requiring the end user to write, understand or interpret code.

“Sensitive” artificial intelligence enables:

• More productive use of expanded (big, often unstructured) information sources

• Intuitive man-machine interactions (no code-speak here!)

• Adaptive, immersive experiences and environments

As frequently touted on the nightly news, AI’s popularity is clear. However, the term’s ubiquity often results in the overestimation or, more and more often, underestimation of what AI can do. To clear the air, let's explore the boundaries of AI's capabilities today.

Why All the Fuss?

Reading, Writing and Arithmetic

Making Sense of AI 1 2 3 4 5 3

ALL AI OR MACHINE LEARNING (ML) driven systems learn from data without being explicitly programmed (that’s the math). But creating a coopera-tive environment in which man and machine work seamlessly together takes more than a smart machine. Deploying smart systems in ways we humans find natural and intuitive is the science and the art.

This means AI must also have the ability to ingest information and/or interact using natural sensory constructs. This may include spoken and written language, visual obser-vations of pictures, video or objects, and so forth. In other words, today's AI solutions should be able to observe, analyze and respond intelligently to their environment in the native tongue.1

1 In the case of machine-to-machine interactions, such as the IoT ecosystem, the “native tongue” may be binary!

Making Sense of AI 1 2 3 4 5 4

Detect, identify and understand the context of objects (people, places and things) in pictures, videos and real life. This includes the ability to translate or interpret text, written language and symbols.

Capture, catalog and interpret spoken commands as well as speech, sounds and auditory signals within the ambient environ-ment or from video.

Communicate insights, directions or responses verbally or in writing using natural language. Including ability to apply appropriate dialects, slang or mimic individual speech patterns. Such as this sentence fragment.

Ingest diverse inputs such as environmental conditions (e.g., temperature, wind, precipitation) or biometrics (e.g., heart rate, perspiration) utilizing sensors and other devices often associated with the internet of things (IoT).

Apply machine learning and other analytic techniques to integrate, analyze and generate insight from input information and signals.

Reading, Writing and Arithmetic

SEEING

HEARING

FEELING

THINKING

SPEAKING/WRITING

roll over for capabilities

Making Sense of AI 1 2 3 4 5 5

MODERN AI APPLICATIONS are smart, adaptive and interactive. But just how smart are we talking? The figure below provides a simple, gross model for gauging the spectrum of machine understanding.

The Holy Grail: From Seeing to Knowing

RECOGNITION COMPREHENSION ABSTRACTION

Simply identifying people or things (objects,

sounds, etc.).

Understanding how things relate; gaining and applying

insight based on the context of a given situation.

Imagining or conceiving new ideas or theo-

ries based not just on discrete, observed inputs or events but on intuition or ‘leaps of logic’ (facts

not in evidence).

We are here.

Making Sense of AI 1 2 3 4 5 6

Great progress has been made improving the machine’s ability to understand very complex, yet discrete, situations or events. However, abstraction remains beyond the reach of AI today.

But, as these simple examples show, achieving full cognition isn’t as simple as it sounds (pun intended!).

1. Vision

Ball

Hoop

Player It's a Ballgame.This team can win the championship.

This player is 28% more likely to receive a 4-year college scholarship to

play ball.

Identify things in a picture or video.

Relate items to each other. Evaluate and predict future performance.

The Holy Grail: From Seeing to Knowing

RECOGNITION COMPREHENSION COMPREHENSION

Making Sense of AI 1 2 3 4 5

2. Interpret Speech

Translate speech verbatim. Recognize this as an exclama-tion, and not someone smoking during a religious experience.

Posit a positive or negative response.

RECOGNITION COMPREHENSION COMPREHENSION

The Holy Grail: From Seeing to Knowing

"Holy Smokes!"

NOT

OR

7

Making Sense of AI 1 2 3

1 Use Facial Recognition to Identify

Patient

2 Ask Patient Triage Questions

3 Record Answers

4 Collect Patient Vitals

5 Analyze Patient Symptoms &

History

6 Prioritize Patient for Triage

7 Add Patient to Triage List & Notify

Nurse

8 Assign Patient to Appropriate

Waiting Area

AI in Practice: Healthcare

3

1

5, 6

2, 8

4, 7

Urgent Care Triage

4 5 8

1

1, 2

3, 4

6

1, 5, 6

Smart Thermostat

1 Monitor Internal & External

Environment

2 Identify Movement

3 Assess Level, Frequency of Activity

4 Analyze Current Conditions &

Prior Energy Settings

5 Adjust Thermostat to Cool Room

6 Send Alert to Homeowner’s App

AI in Practice: Energy

Making Sense of AI 1 2 3 4 5 9

1, 2

2, 3, 4

5

6, 7

Production Line Quality Control

1 Observe Chip on Production Line

2 Analyze Image for Microscopic

Anomalies

3 Identify Potential Anomaly

4 Assess Potential Impact

5 Sound Critical Alert to Production

Engineer

6 Halt Production Line

7 Log Defect

AI in Practice: Manufacturing

Making Sense of AI 1 2 3 104 5

AI in Practice: Other Applications

Making Sense of AI 1 2 3 114 5

roll over titles for more information

Making Sense of AI 1 2 3 4 5

ARTIFICIAL INTELLIGENCE

solutions are extremely powerful

but not without limitation.

Keep in mind:

Points to Ponder

12

Intuitive, insightful interac-tion is a keystone of AI. Have a question? Just ask. The machine will answer. Out loud. In English (or your preferred language, as determined by the algorithm).

From this perspective, AI systems are designed to emulate how humans commu-nicate. This doesn’t mean that how an intended task or outcome is accomplished can or should be modeled against human practice or process.

AI solutions are smart, not spontaneously creative.

AI solutions become more adept at identi-fying the right action or predicting outcomes as they test and learn from experience (aka, data).

AI will not ideate, project new futures or create net new responses — regardless of the data consumed.

AI Talks but Doesn’t Walk like a Person.

AI Will Adapt but Won’t Create. AI Is Focused.

Practical business solutions are typically specialized: automating or augmenting discrete, well-bounded systemic interactions or points of engagement.

A solution to recom-mend treatment for breast cancer will not apply to colon cancer. Or diabetes.

The personal assis-tant managing your calendar won’t opti-mize your financial portfolio.

Making Sense of AI 1 2 3 13

While AI use cases tend to focus on human- machine interactions AI is also core to machine-to-machine interactions (often via the IoT).

In either case, key inter-section points, rules of engagement and expected outcomes must be deliberately designed.

When it comes to AI, the experience is the product. That experi-ence can’t be prefabri-cated.

Chatbots, for example, must be trained to respond to specific types of inquiries with the responses and tone that reflect your offer-ings and brand.

Conversational design (which is not limited to just text or verbal inter-action) is the new user experience.

AI Doesn’t Discriminate. (between man and machine)

AI may automate discrete tasks or interactions, augment existing decision pathways or extend them.

Creating appropriate, easily consumable interactions between man and machine requires clarity about the scope of the problem being solved and the intended expe-rience.

AI Is an Experience.AI Automates, Augments or Extends.

Points to PonderARTIFICIAL

INTELLIGENCE solutions are

extremely powerful but not without

limitation.

Keep in mind:

4 5

About the AuthorKIMBERLY NEVALA is the Director of Business Strategies for SAS Best Practices. She is responsible for market analysis, industry education, emerging best practices and strategies in the areas of advanced analytics, information governance and data-driven culture. Kimberly’s current focus is helping clients understand the business potential as well as the practical impli-cations and limitations of artificial intelligence and machine learning.

A popular speaker and author, Kimberly has published numerous e-books and white papers including: The Machine Learning Primer, Portrait of a CAO, Data-Driven: Going from All Gut to Analytically-Driven Glory, The Anatomy of an Analytic Enterprise, Sustainable Data Governance and Top 10 Mistakes to Avoid When Launching A Data Governance Program.

©2018 SAS Institute Inc. All rights reserved.

SAS Institute Inc.100 SAS Campus DriveCary, NC 27513-2414, USAPhone: 919-677-8000Email: bestpractices@SAS.com

Learn MoreFor more insight into AI visit: www.sas.com/artificialintelligence

SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies. 109303_G68731.0118